git.sesse.net Git - bcachefs-tools-debian/blob - libbcachefs/recovery.c

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include "bcachefs.h"
   4 #include "bkey_buf.h"
   5 #include "alloc_background.h"
   6 #include "btree_gc.h"
   7 #include "btree_update.h"
   8 #include "btree_update_interior.h"
   9 #include "btree_io.h"
  10 #include "buckets.h"
  11 #include "dirent.h"
  12 #include "ec.h"
  13 #include "error.h"
  14 #include "fs-common.h"
  15 #include "fsck.h"
  16 #include "journal_io.h"
  17 #include "journal_reclaim.h"
  18 #include "journal_seq_blacklist.h"
  19 #include "move.h"
  20 #include "quota.h"
  21 #include "recovery.h"
  22 #include "replicas.h"
  23 #include "subvolume.h"
  24 #include "super-io.h"
  25
  26 #include <linux/sort.h>
  27 #include <linux/stat.h>
  28
  29 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
  30
  31 /* for -o reconstruct_alloc: */
  32 static void drop_alloc_keys(struct journal_keys *keys)
  33 {
  34         size_t src, dst;
  35
  36         for (src = 0, dst = 0; src < keys->nr; src++)
  37                 if (keys->d[src].btree_id != BTREE_ID_alloc)
  38                         keys->d[dst++] = keys->d[src];
  39
  40         keys->nr = dst;
  41 }
  42
  43 /*
  44  * Btree node pointers have a field to stack a pointer to the in memory btree
  45  * node; we need to zero out this field when reading in btree nodes, or when
  46  * reading in keys from the journal:
  47  */
  48 static void zero_out_btree_mem_ptr(struct journal_keys *keys)
  49 {
  50         struct journal_key *i;
  51
  52         for (i = keys->d; i < keys->d + keys->nr; i++)
  53                 if (i->k->k.type == KEY_TYPE_btree_ptr_v2)
  54                         bkey_i_to_btree_ptr_v2(i->k)->v.mem_ptr = 0;
  55 }
  56
  57 /* iterate over keys read from the journal: */
  58
  59 static int __journal_key_cmp(enum btree_id      l_btree_id,
  60                              unsigned           l_level,
  61                              struct bpos        l_pos,
  62                              struct journal_key *r)
  63 {
  64         return (cmp_int(l_btree_id,     r->btree_id) ?:
  65                 cmp_int(l_level,        r->level) ?:
  66                 bpos_cmp(l_pos, r->k->k.p));
  67 }
  68
  69 static int journal_key_cmp(struct journal_key *l, struct journal_key *r)
  70 {
  71         return (cmp_int(l->btree_id,    r->btree_id) ?:
  72                 cmp_int(l->level,       r->level) ?:
  73                 bpos_cmp(l->k->k.p,     r->k->k.p));
  74 }
  75
  76 static size_t journal_key_search(struct journal_keys *journal_keys,
  77                                  enum btree_id id, unsigned level,
  78                                  struct bpos pos)
  79 {
  80         size_t l = 0, r = journal_keys->nr, m;
  81
  82         while (l < r) {
  83                 m = l + ((r - l) >> 1);
  84                 if (__journal_key_cmp(id, level, pos, &journal_keys->d[m]) > 0)
  85                         l = m + 1;
  86                 else
  87                         r = m;
  88         }
  89
  90         BUG_ON(l < journal_keys->nr &&
  91                __journal_key_cmp(id, level, pos, &journal_keys->d[l]) > 0);
  92
  93         BUG_ON(l &&
  94                __journal_key_cmp(id, level, pos, &journal_keys->d[l - 1]) <= 0);
  95
  96         return l;
  97 }
  98
  99 static void journal_iter_fix(struct bch_fs *c, struct journal_iter *iter, unsigned idx)
 100 {
 101         struct bkey_i *n = iter->keys->d[idx].k;
 102         struct btree_and_journal_iter *biter =
 103                 container_of(iter, struct btree_and_journal_iter, journal);
 104
 105         if (iter->idx > idx ||
 106             (iter->idx == idx &&
 107              biter->last &&
 108              bpos_cmp(n->k.p, biter->unpacked.p) <= 0))
 109                 iter->idx++;
 110 }
 111
 112 int bch2_journal_key_insert_take(struct bch_fs *c, enum btree_id id,
 113                                  unsigned level, struct bkey_i *k)
 114 {
 115         struct journal_key n = {
 116                 .btree_id       = id,
 117                 .level          = level,
 118                 .k              = k,
 119                 .allocated      = true
 120         };
 121         struct journal_keys *keys = &c->journal_keys;
 122         struct journal_iter *iter;
 123         unsigned idx = journal_key_search(keys, id, level, k->k.p);
 124
 125         if (idx < keys->nr &&
 126             journal_key_cmp(&n, &keys->d[idx]) == 0) {
 127                 if (keys->d[idx].allocated)
 128                         kfree(keys->d[idx].k);
 129                 keys->d[idx] = n;
 130                 return 0;
 131         }
 132
 133         if (keys->nr == keys->size) {
 134                 struct journal_keys new_keys = {
 135                         .nr                     = keys->nr,
 136                         .size                   = keys->size * 2,
 137                         .journal_seq_base       = keys->journal_seq_base,
 138                 };
 139
 140                 new_keys.d = kvmalloc(sizeof(new_keys.d[0]) * new_keys.size, GFP_KERNEL);
 141                 if (!new_keys.d) {
 142                         bch_err(c, "%s: error allocating new key array (size %zu)",
 143                                 __func__, new_keys.size);
 144                         return -ENOMEM;
 145                 }
 146
 147                 memcpy(new_keys.d, keys->d, sizeof(keys->d[0]) * keys->nr);
 148                 kvfree(keys->d);
 149                 *keys = new_keys;
 150         }
 151
 152         array_insert_item(keys->d, keys->nr, idx, n);
 153
 154         list_for_each_entry(iter, &c->journal_iters, list)
 155                 journal_iter_fix(c, iter, idx);
 156
 157         return 0;
 158 }
 159
 160 int bch2_journal_key_insert(struct bch_fs *c, enum btree_id id,
 161                             unsigned level, struct bkey_i *k)
 162 {
 163         struct bkey_i *n;
 164         int ret;
 165
 166         n = kmalloc(bkey_bytes(&k->k), GFP_KERNEL);
 167         if (!n)
 168                 return -ENOMEM;
 169
 170         bkey_copy(n, k);
 171         ret = bch2_journal_key_insert_take(c, id, level, n);
 172         if (ret)
 173                 kfree(n);
 174         return ret;
 175 }
 176
 177 int bch2_journal_key_delete(struct bch_fs *c, enum btree_id id,
 178                             unsigned level, struct bpos pos)
 179 {
 180         struct bkey_i whiteout;
 181
 182         bkey_init(&whiteout.k);
 183         whiteout.k.p = pos;
 184
 185         return bch2_journal_key_insert(c, id, level, &whiteout);
 186 }
 187
 188 static struct bkey_i *bch2_journal_iter_peek(struct journal_iter *iter)
 189 {
 190         struct journal_key *k = iter->idx - iter->keys->nr
 191                 ? iter->keys->d + iter->idx : NULL;
 192
 193         if (k &&
 194             k->btree_id == iter->btree_id &&
 195             k->level    == iter->level)
 196                 return k->k;
 197
 198         iter->idx = iter->keys->nr;
 199         return NULL;
 200 }
 201
 202 static void bch2_journal_iter_advance(struct journal_iter *iter)
 203 {
 204         if (iter->idx < iter->keys->nr)
 205                 iter->idx++;
 206 }
 207
 208 static void bch2_journal_iter_exit(struct journal_iter *iter)
 209 {
 210         list_del(&iter->list);
 211 }
 212
 213 static void bch2_journal_iter_init(struct bch_fs *c,
 214                                    struct journal_iter *iter,
 215                                    enum btree_id id, unsigned level,
 216                                    struct bpos pos)
 217 {
 218         iter->btree_id  = id;
 219         iter->level     = level;
 220         iter->keys      = &c->journal_keys;
 221         iter->idx       = journal_key_search(&c->journal_keys, id, level, pos);
 222         list_add(&iter->list, &c->journal_iters);
 223 }
 224
 225 static struct bkey_s_c bch2_journal_iter_peek_btree(struct btree_and_journal_iter *iter)
 226 {
 227         return bch2_btree_node_iter_peek_unpack(&iter->node_iter,
 228                                                 iter->b, &iter->unpacked);
 229 }
 230
 231 static void bch2_journal_iter_advance_btree(struct btree_and_journal_iter *iter)
 232 {
 233         bch2_btree_node_iter_advance(&iter->node_iter, iter->b);
 234 }
 235
 236 void bch2_btree_and_journal_iter_advance(struct btree_and_journal_iter *iter)
 237 {
 238         switch (iter->last) {
 239         case none:
 240                 break;
 241         case btree:
 242                 bch2_journal_iter_advance_btree(iter);
 243                 break;
 244         case journal:
 245                 bch2_journal_iter_advance(&iter->journal);
 246                 break;
 247         }
 248
 249         iter->last = none;
 250 }
 251
 252 struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *iter)
 253 {
 254         struct bkey_s_c ret;
 255
 256         while (1) {
 257                 struct bkey_s_c btree_k         =
 258                         bch2_journal_iter_peek_btree(iter);
 259                 struct bkey_s_c journal_k       =
 260                         bkey_i_to_s_c(bch2_journal_iter_peek(&iter->journal));
 261
 262                 if (btree_k.k && journal_k.k) {
 263                         int cmp = bpos_cmp(btree_k.k->p, journal_k.k->p);
 264
 265                         if (!cmp)
 266                                 bch2_journal_iter_advance_btree(iter);
 267
 268                         iter->last = cmp < 0 ? btree : journal;
 269                 } else if (btree_k.k) {
 270                         iter->last = btree;
 271                 } else if (journal_k.k) {
 272                         iter->last = journal;
 273                 } else {
 274                         iter->last = none;
 275                         return bkey_s_c_null;
 276                 }
 277
 278                 ret = iter->last == journal ? journal_k : btree_k;
 279
 280                 if (iter->b &&
 281                     bpos_cmp(ret.k->p, iter->b->data->max_key) > 0) {
 282                         iter->journal.idx = iter->journal.keys->nr;
 283                         iter->last = none;
 284                         return bkey_s_c_null;
 285                 }
 286
 287                 if (!bkey_deleted(ret.k))
 288                         break;
 289
 290                 bch2_btree_and_journal_iter_advance(iter);
 291         }
 292
 293         return ret;
 294 }
 295
 296 struct bkey_s_c bch2_btree_and_journal_iter_next(struct btree_and_journal_iter *iter)
 297 {
 298         bch2_btree_and_journal_iter_advance(iter);
 299
 300         return bch2_btree_and_journal_iter_peek(iter);
 301 }
 302
 303 void bch2_btree_and_journal_iter_exit(struct btree_and_journal_iter *iter)
 304 {
 305         bch2_journal_iter_exit(&iter->journal);
 306 }
 307
 308 void bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
 309                                                 struct bch_fs *c,
 310                                                 struct btree *b)
 311 {
 312         memset(iter, 0, sizeof(*iter));
 313
 314         iter->b = b;
 315         bch2_btree_node_iter_init_from_start(&iter->node_iter, iter->b);
 316         bch2_journal_iter_init(c, &iter->journal,
 317                                b->c.btree_id, b->c.level, b->data->min_key);
 318 }
 319
 320 /* Walk btree, overlaying keys from the journal: */
 321
 322 static void btree_and_journal_iter_prefetch(struct bch_fs *c, struct btree *b,
 323                                            struct btree_and_journal_iter iter)
 324 {
 325         unsigned i = 0, nr = b->c.level > 1 ? 2 : 16;
 326         struct bkey_s_c k;
 327         struct bkey_buf tmp;
 328
 329         BUG_ON(!b->c.level);
 330
 331         bch2_bkey_buf_init(&tmp);
 332
 333         while (i < nr &&
 334                (k = bch2_btree_and_journal_iter_peek(&iter)).k) {
 335                 bch2_bkey_buf_reassemble(&tmp, c, k);
 336
 337                 bch2_btree_node_prefetch(c, NULL, NULL, tmp.k,
 338                                         b->c.btree_id, b->c.level - 1);
 339
 340                 bch2_btree_and_journal_iter_advance(&iter);
 341                 i++;
 342         }
 343
 344         bch2_bkey_buf_exit(&tmp, c);
 345 }
 346
 347 static int bch2_btree_and_journal_walk_recurse(struct btree_trans *trans, struct btree *b,
 348                                 enum btree_id btree_id,
 349                                 btree_walk_key_fn key_fn)
 350 {
 351         struct bch_fs *c = trans->c;
 352         struct btree_and_journal_iter iter;
 353         struct bkey_s_c k;
 354         struct bkey_buf tmp;
 355         struct btree *child;
 356         int ret = 0;
 357
 358         bch2_bkey_buf_init(&tmp);
 359         bch2_btree_and_journal_iter_init_node_iter(&iter, c, b);
 360
 361         while ((k = bch2_btree_and_journal_iter_peek(&iter)).k) {
 362                 if (b->c.level) {
 363                         bch2_bkey_buf_reassemble(&tmp, c, k);
 364
 365                         child = bch2_btree_node_get_noiter(c, tmp.k,
 366                                                 b->c.btree_id, b->c.level - 1,
 367                                                 false);
 368
 369                         ret = PTR_ERR_OR_ZERO(child);
 370                         if (ret)
 371                                 break;
 372
 373                         btree_and_journal_iter_prefetch(c, b, iter);
 374
 375                         ret = bch2_btree_and_journal_walk_recurse(trans, child,
 376                                         btree_id, key_fn);
 377                         six_unlock_read(&child->c.lock);
 378                 } else {
 379                         ret = key_fn(trans, k);
 380                 }
 381
 382                 if (ret)
 383                         break;
 384
 385                 bch2_btree_and_journal_iter_advance(&iter);
 386         }
 387
 388         bch2_btree_and_journal_iter_exit(&iter);
 389         bch2_bkey_buf_exit(&tmp, c);
 390         return ret;
 391 }
 392
 393 int bch2_btree_and_journal_walk(struct btree_trans *trans, enum btree_id btree_id,
 394                                 btree_walk_key_fn key_fn)
 395 {
 396         struct bch_fs *c = trans->c;
 397         struct btree *b = c->btree_roots[btree_id].b;
 398         int ret = 0;
 399
 400         if (btree_node_fake(b))
 401                 return 0;
 402
 403         six_lock_read(&b->c.lock, NULL, NULL);
 404         ret = bch2_btree_and_journal_walk_recurse(trans, b, btree_id, key_fn);
 405         six_unlock_read(&b->c.lock);
 406
 407         return ret;
 408 }
 409
 410 /* sort and dedup all keys in the journal: */
 411
 412 void bch2_journal_entries_free(struct list_head *list)
 413 {
 414
 415         while (!list_empty(list)) {
 416                 struct journal_replay *i =
 417                         list_first_entry(list, struct journal_replay, list);
 418                 list_del(&i->list);
 419                 kvpfree(i, offsetof(struct journal_replay, j) +
 420                         vstruct_bytes(&i->j));
 421         }
 422 }
 423
 424 /*
 425  * When keys compare equal, oldest compares first:
 426  */
 427 static int journal_sort_key_cmp(const void *_l, const void *_r)
 428 {
 429         const struct journal_key *l = _l;
 430         const struct journal_key *r = _r;
 431
 432         return  cmp_int(l->btree_id,    r->btree_id) ?:
 433                 cmp_int(l->level,       r->level) ?:
 434                 bpos_cmp(l->k->k.p, r->k->k.p) ?:
 435                 cmp_int(l->journal_seq, r->journal_seq) ?:
 436                 cmp_int(l->journal_offset, r->journal_offset);
 437 }
 438
 439 void bch2_journal_keys_free(struct journal_keys *keys)
 440 {
 441         struct journal_key *i;
 442
 443         for (i = keys->d; i < keys->d + keys->nr; i++)
 444                 if (i->allocated)
 445                         kfree(i->k);
 446
 447         kvfree(keys->d);
 448         keys->d = NULL;
 449         keys->nr = 0;
 450 }
 451
 452 static struct journal_keys journal_keys_sort(struct list_head *journal_entries)
 453 {
 454         struct journal_replay *i;
 455         struct jset_entry *entry;
 456         struct bkey_i *k, *_n;
 457         struct journal_keys keys = { NULL };
 458         struct journal_key *src, *dst;
 459         size_t nr_keys = 0;
 460
 461         if (list_empty(journal_entries))
 462                 return keys;
 463
 464         list_for_each_entry(i, journal_entries, list) {
 465                 if (i->ignore)
 466                         continue;
 467
 468                 if (!keys.journal_seq_base)
 469                         keys.journal_seq_base = le64_to_cpu(i->j.seq);
 470
 471                 for_each_jset_key(k, _n, entry, &i->j)
 472                         nr_keys++;
 473         }
 474
 475         keys.size = roundup_pow_of_two(nr_keys);
 476
 477         keys.d = kvmalloc(sizeof(keys.d[0]) * keys.size, GFP_KERNEL);
 478         if (!keys.d)
 479                 goto err;
 480
 481         list_for_each_entry(i, journal_entries, list) {
 482                 if (i->ignore)
 483                         continue;
 484
 485                 BUG_ON(le64_to_cpu(i->j.seq) - keys.journal_seq_base > U32_MAX);
 486
 487                 for_each_jset_key(k, _n, entry, &i->j)
 488                         keys.d[keys.nr++] = (struct journal_key) {
 489                                 .btree_id       = entry->btree_id,
 490                                 .level          = entry->level,
 491                                 .k              = k,
 492                                 .journal_seq    = le64_to_cpu(i->j.seq) -
 493                                         keys.journal_seq_base,
 494                                 .journal_offset = k->_data - i->j._data,
 495                         };
 496         }
 497
 498         sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_key_cmp, NULL);
 499
 500         src = dst = keys.d;
 501         while (src < keys.d + keys.nr) {
 502                 while (src + 1 < keys.d + keys.nr &&
 503                        src[0].btree_id  == src[1].btree_id &&
 504                        src[0].level     == src[1].level &&
 505                        !bpos_cmp(src[0].k->k.p, src[1].k->k.p))
 506                         src++;
 507
 508                 *dst++ = *src++;
 509         }
 510
 511         keys.nr = dst - keys.d;
 512 err:
 513         return keys;
 514 }
 515
 516 /* journal replay: */
 517
 518 static void replay_now_at(struct journal *j, u64 seq)
 519 {
 520         BUG_ON(seq < j->replay_journal_seq);
 521         BUG_ON(seq > j->replay_journal_seq_end);
 522
 523         while (j->replay_journal_seq < seq)
 524                 bch2_journal_pin_put(j, j->replay_journal_seq++);
 525 }
 526
 527 static int __bch2_journal_replay_key(struct btree_trans *trans,
 528                                      struct journal_key *k)
 529 {
 530         struct btree_iter iter;
 531         unsigned iter_flags =
 532                 BTREE_ITER_INTENT|
 533                 BTREE_ITER_NOT_EXTENTS;
 534         int ret;
 535
 536         if (!k->level && k->btree_id == BTREE_ID_alloc)
 537                 iter_flags |= BTREE_ITER_CACHED|BTREE_ITER_CACHED_NOFILL;
 538
 539         bch2_trans_node_iter_init(trans, &iter, k->btree_id, k->k->k.p,
 540                                   BTREE_MAX_DEPTH, k->level,
 541                                   iter_flags);
 542         ret   = bch2_btree_iter_traverse(&iter) ?:
 543                 bch2_trans_update(trans, &iter, k->k, BTREE_TRIGGER_NORUN);
 544         bch2_trans_iter_exit(trans, &iter);
 545         return ret;
 546 }
 547
 548 static int bch2_journal_replay_key(struct bch_fs *c, struct journal_key *k)
 549 {
 550         unsigned commit_flags =
 551                 BTREE_INSERT_LAZY_RW|
 552                 BTREE_INSERT_NOFAIL|
 553                 BTREE_INSERT_JOURNAL_RESERVED;
 554
 555         if (!k->allocated)
 556                 commit_flags |= BTREE_INSERT_JOURNAL_REPLAY;
 557
 558         return bch2_trans_do(c, NULL, NULL, commit_flags,
 559                              __bch2_journal_replay_key(&trans, k));
 560 }
 561
 562 static int journal_sort_seq_cmp(const void *_l, const void *_r)
 563 {
 564         const struct journal_key *l = *((const struct journal_key **)_l);
 565         const struct journal_key *r = *((const struct journal_key **)_r);
 566
 567         return  cmp_int(r->level,       l->level) ?:
 568                 cmp_int(l->journal_seq, r->journal_seq) ?:
 569                 cmp_int(l->btree_id,    r->btree_id) ?:
 570                 bpos_cmp(l->k->k.p,     r->k->k.p);
 571 }
 572
 573 static int bch2_journal_replay(struct bch_fs *c)
 574 {
 575         struct journal_keys *keys = &c->journal_keys;
 576         struct journal_key **keys_sorted, *k;
 577         struct journal *j = &c->journal;
 578         struct bch_dev *ca;
 579         unsigned idx;
 580         size_t i;
 581         u64 seq;
 582         int ret;
 583
 584         keys_sorted = kmalloc_array(sizeof(*keys_sorted), keys->nr, GFP_KERNEL);
 585         if (!keys_sorted)
 586                 return -ENOMEM;
 587
 588         for (i = 0; i < keys->nr; i++)
 589                 keys_sorted[i] = &keys->d[i];
 590
 591         sort(keys_sorted, keys->nr,
 592              sizeof(keys_sorted[0]),
 593              journal_sort_seq_cmp, NULL);
 594
 595         if (keys->nr)
 596                 replay_now_at(j, keys->journal_seq_base);
 597
 598         seq = j->replay_journal_seq;
 599
 600         /*
 601          * First replay updates to the alloc btree - these will only update the
 602          * btree key cache:
 603          */
 604         for (i = 0; i < keys->nr; i++) {
 605                 k = keys_sorted[i];
 606
 607                 cond_resched();
 608
 609                 if (!k->level && k->btree_id == BTREE_ID_alloc) {
 610                         j->replay_journal_seq = keys->journal_seq_base + k->journal_seq;
 611                         ret = bch2_journal_replay_key(c, k);
 612                         if (ret)
 613                                 goto err;
 614                 }
 615         }
 616
 617         /* Now we can start the allocator threads: */
 618         set_bit(BCH_FS_ALLOC_REPLAY_DONE, &c->flags);
 619         for_each_member_device(ca, c, idx)
 620                 bch2_wake_allocator(ca);
 621
 622         /*
 623          * Next replay updates to interior btree nodes:
 624          */
 625         for (i = 0; i < keys->nr; i++) {
 626                 k = keys_sorted[i];
 627
 628                 cond_resched();
 629
 630                 if (k->level) {
 631                         j->replay_journal_seq = keys->journal_seq_base + k->journal_seq;
 632                         ret = bch2_journal_replay_key(c, k);
 633                         if (ret)
 634                                 goto err;
 635                 }
 636         }
 637
 638         /*
 639          * Now that the btree is in a consistent state, we can start journal
 640          * reclaim (which will be flushing entries from the btree key cache back
 641          * to the btree:
 642          */
 643         set_bit(BCH_FS_BTREE_INTERIOR_REPLAY_DONE, &c->flags);
 644         set_bit(JOURNAL_RECLAIM_STARTED, &j->flags);
 645         journal_reclaim_kick(j);
 646
 647         j->replay_journal_seq = seq;
 648
 649         /*
 650          * Now replay leaf node updates:
 651          */
 652         for (i = 0; i < keys->nr; i++) {
 653                 k = keys_sorted[i];
 654
 655                 cond_resched();
 656
 657                 if (k->level || k->btree_id == BTREE_ID_alloc)
 658                         continue;
 659
 660                 replay_now_at(j, keys->journal_seq_base + k->journal_seq);
 661
 662                 ret = bch2_journal_replay_key(c, k);
 663                 if (ret)
 664                         goto err;
 665         }
 666
 667         replay_now_at(j, j->replay_journal_seq_end);
 668         j->replay_journal_seq = 0;
 669
 670         bch2_journal_set_replay_done(j);
 671         bch2_journal_flush_all_pins(j);
 672         kfree(keys_sorted);
 673
 674         return bch2_journal_error(j);
 675 err:
 676         bch_err(c, "journal replay: error %d while replaying key at btree %s level %u",
 677                 ret, bch2_btree_ids[k->btree_id], k->level);
 678         kfree(keys_sorted);
 679
 680         return ret;
 681 }
 682
 683 /* journal replay early: */
 684
 685 static int journal_replay_entry_early(struct bch_fs *c,
 686                                       struct jset_entry *entry)
 687 {
 688         int ret = 0;
 689
 690         switch (entry->type) {
 691         case BCH_JSET_ENTRY_btree_root: {
 692                 struct btree_root *r;
 693
 694                 if (entry->btree_id >= BTREE_ID_NR) {
 695                         bch_err(c, "filesystem has unknown btree type %u",
 696                                 entry->btree_id);
 697                         return -EINVAL;
 698                 }
 699
 700                 r = &c->btree_roots[entry->btree_id];
 701
 702                 if (entry->u64s) {
 703                         r->level = entry->level;
 704                         bkey_copy(&r->key, &entry->start[0]);
 705                         r->error = 0;
 706                 } else {
 707                         r->error = -EIO;
 708                 }
 709                 r->alive = true;
 710                 break;
 711         }
 712         case BCH_JSET_ENTRY_usage: {
 713                 struct jset_entry_usage *u =
 714                         container_of(entry, struct jset_entry_usage, entry);
 715
 716                 switch (entry->btree_id) {
 717                 case BCH_FS_USAGE_reserved:
 718                         if (entry->level < BCH_REPLICAS_MAX)
 719                                 c->usage_base->persistent_reserved[entry->level] =
 720                                         le64_to_cpu(u->v);
 721                         break;
 722                 case BCH_FS_USAGE_inodes:
 723                         c->usage_base->nr_inodes = le64_to_cpu(u->v);
 724                         break;
 725                 case BCH_FS_USAGE_key_version:
 726                         atomic64_set(&c->key_version,
 727                                      le64_to_cpu(u->v));
 728                         break;
 729                 }
 730
 731                 break;
 732         }
 733         case BCH_JSET_ENTRY_data_usage: {
 734                 struct jset_entry_data_usage *u =
 735                         container_of(entry, struct jset_entry_data_usage, entry);
 736
 737                 ret = bch2_replicas_set_usage(c, &u->r,
 738                                               le64_to_cpu(u->v));
 739                 break;
 740         }
 741         case BCH_JSET_ENTRY_dev_usage: {
 742                 struct jset_entry_dev_usage *u =
 743                         container_of(entry, struct jset_entry_dev_usage, entry);
 744                 struct bch_dev *ca = bch_dev_bkey_exists(c, le32_to_cpu(u->dev));
 745                 unsigned i, nr_types = jset_entry_dev_usage_nr_types(u);
 746
 747                 ca->usage_base->buckets_ec              = le64_to_cpu(u->buckets_ec);
 748                 ca->usage_base->buckets_unavailable     = le64_to_cpu(u->buckets_unavailable);
 749
 750                 for (i = 0; i < min_t(unsigned, nr_types, BCH_DATA_NR); i++) {
 751                         ca->usage_base->d[i].buckets    = le64_to_cpu(u->d[i].buckets);
 752                         ca->usage_base->d[i].sectors    = le64_to_cpu(u->d[i].sectors);
 753                         ca->usage_base->d[i].fragmented = le64_to_cpu(u->d[i].fragmented);
 754                 }
 755
 756                 break;
 757         }
 758         case BCH_JSET_ENTRY_blacklist: {
 759                 struct jset_entry_blacklist *bl_entry =
 760                         container_of(entry, struct jset_entry_blacklist, entry);
 761
 762                 ret = bch2_journal_seq_blacklist_add(c,
 763                                 le64_to_cpu(bl_entry->seq),
 764                                 le64_to_cpu(bl_entry->seq) + 1);
 765                 break;
 766         }
 767         case BCH_JSET_ENTRY_blacklist_v2: {
 768                 struct jset_entry_blacklist_v2 *bl_entry =
 769                         container_of(entry, struct jset_entry_blacklist_v2, entry);
 770
 771                 ret = bch2_journal_seq_blacklist_add(c,
 772                                 le64_to_cpu(bl_entry->start),
 773                                 le64_to_cpu(bl_entry->end) + 1);
 774                 break;
 775         }
 776         case BCH_JSET_ENTRY_clock: {
 777                 struct jset_entry_clock *clock =
 778                         container_of(entry, struct jset_entry_clock, entry);
 779
 780                 atomic64_set(&c->io_clock[clock->rw].now, le64_to_cpu(clock->time));
 781         }
 782         }
 783
 784         return ret;
 785 }
 786
 787 static int journal_replay_early(struct bch_fs *c,
 788                                 struct bch_sb_field_clean *clean,
 789                                 struct list_head *journal)
 790 {
 791         struct journal_replay *i;
 792         struct jset_entry *entry;
 793         int ret;
 794
 795         if (clean) {
 796                 for (entry = clean->start;
 797                      entry != vstruct_end(&clean->field);
 798                      entry = vstruct_next(entry)) {
 799                         ret = journal_replay_entry_early(c, entry);
 800                         if (ret)
 801                                 return ret;
 802                 }
 803         } else {
 804                 list_for_each_entry(i, journal, list) {
 805                         if (i->ignore)
 806                                 continue;
 807
 808                         vstruct_for_each(&i->j, entry) {
 809                                 ret = journal_replay_entry_early(c, entry);
 810                                 if (ret)
 811                                         return ret;
 812                         }
 813                 }
 814         }
 815
 816         bch2_fs_usage_initialize(c);
 817
 818         return 0;
 819 }
 820
 821 /* sb clean section: */
 822
 823 static struct bkey_i *btree_root_find(struct bch_fs *c,
 824                                       struct bch_sb_field_clean *clean,
 825                                       struct jset *j,
 826                                       enum btree_id id, unsigned *level)
 827 {
 828         struct bkey_i *k;
 829         struct jset_entry *entry, *start, *end;
 830
 831         if (clean) {
 832                 start = clean->start;
 833                 end = vstruct_end(&clean->field);
 834         } else {
 835                 start = j->start;
 836                 end = vstruct_last(j);
 837         }
 838
 839         for (entry = start; entry < end; entry = vstruct_next(entry))
 840                 if (entry->type == BCH_JSET_ENTRY_btree_root &&
 841                     entry->btree_id == id)
 842                         goto found;
 843
 844         return NULL;
 845 found:
 846         if (!entry->u64s)
 847                 return ERR_PTR(-EINVAL);
 848
 849         k = entry->start;
 850         *level = entry->level;
 851         return k;
 852 }
 853
 854 static int verify_superblock_clean(struct bch_fs *c,
 855                                    struct bch_sb_field_clean **cleanp,
 856                                    struct jset *j)
 857 {
 858         unsigned i;
 859         struct bch_sb_field_clean *clean = *cleanp;
 860         int ret = 0;
 861
 862         if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
 863                         "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
 864                         le64_to_cpu(clean->journal_seq),
 865                         le64_to_cpu(j->seq))) {
 866                 kfree(clean);
 867                 *cleanp = NULL;
 868                 return 0;
 869         }
 870
 871         for (i = 0; i < BTREE_ID_NR; i++) {
 872                 char buf1[200], buf2[200];
 873                 struct bkey_i *k1, *k2;
 874                 unsigned l1 = 0, l2 = 0;
 875
 876                 k1 = btree_root_find(c, clean, NULL, i, &l1);
 877                 k2 = btree_root_find(c, NULL, j, i, &l2);
 878
 879                 if (!k1 && !k2)
 880                         continue;
 881
 882                 mustfix_fsck_err_on(!k1 || !k2 ||
 883                                     IS_ERR(k1) ||
 884                                     IS_ERR(k2) ||
 885                                     k1->k.u64s != k2->k.u64s ||
 886                                     memcmp(k1, k2, bkey_bytes(k1)) ||
 887                                     l1 != l2, c,
 888                         "superblock btree root %u doesn't match journal after clean shutdown\n"
 889                         "sb:      l=%u %s\n"
 890                         "journal: l=%u %s\n", i,
 891                         l1, (bch2_bkey_val_to_text(&PBUF(buf1), c, bkey_i_to_s_c(k1)), buf1),
 892                         l2, (bch2_bkey_val_to_text(&PBUF(buf2), c, bkey_i_to_s_c(k2)), buf2));
 893         }
 894 fsck_err:
 895         return ret;
 896 }
 897
 898 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
 899 {
 900         struct bch_sb_field_clean *clean, *sb_clean;
 901         int ret;
 902
 903         mutex_lock(&c->sb_lock);
 904         sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
 905
 906         if (fsck_err_on(!sb_clean, c,
 907                         "superblock marked clean but clean section not present")) {
 908                 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 909                 c->sb.clean = false;
 910                 mutex_unlock(&c->sb_lock);
 911                 return NULL;
 912         }
 913
 914         clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
 915                         GFP_KERNEL);
 916         if (!clean) {
 917                 mutex_unlock(&c->sb_lock);
 918                 return ERR_PTR(-ENOMEM);
 919         }
 920
 921         ret = bch2_sb_clean_validate(c, clean, READ);
 922         if (ret) {
 923                 mutex_unlock(&c->sb_lock);
 924                 return ERR_PTR(ret);
 925         }
 926
 927         mutex_unlock(&c->sb_lock);
 928
 929         return clean;
 930 fsck_err:
 931         mutex_unlock(&c->sb_lock);
 932         return ERR_PTR(ret);
 933 }
 934
 935 static int read_btree_roots(struct bch_fs *c)
 936 {
 937         unsigned i;
 938         int ret = 0;
 939
 940         for (i = 0; i < BTREE_ID_NR; i++) {
 941                 struct btree_root *r = &c->btree_roots[i];
 942
 943                 if (!r->alive)
 944                         continue;
 945
 946                 if (i == BTREE_ID_alloc &&
 947                     c->opts.reconstruct_alloc) {
 948                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 949                         continue;
 950                 }
 951
 952                 if (r->error) {
 953                         __fsck_err(c, i == BTREE_ID_alloc
 954                                    ? FSCK_CAN_IGNORE : 0,
 955                                    "invalid btree root %s",
 956                                    bch2_btree_ids[i]);
 957                         if (i == BTREE_ID_alloc)
 958                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 959                 }
 960
 961                 ret = bch2_btree_root_read(c, i, &r->key, r->level);
 962                 if (ret) {
 963                         __fsck_err(c, i == BTREE_ID_alloc
 964                                    ? FSCK_CAN_IGNORE : 0,
 965                                    "error reading btree root %s",
 966                                    bch2_btree_ids[i]);
 967                         if (i == BTREE_ID_alloc)
 968                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 969                 }
 970         }
 971
 972         for (i = 0; i < BTREE_ID_NR; i++)
 973                 if (!c->btree_roots[i].b)
 974                         bch2_btree_root_alloc(c, i);
 975 fsck_err:
 976         return ret;
 977 }
 978
 979 static int bch2_fs_initialize_subvolumes(struct bch_fs *c)
 980 {
 981         struct bkey_i_snapshot  root_snapshot;
 982         struct bkey_i_subvolume root_volume;
 983         int ret;
 984
 985         bkey_snapshot_init(&root_snapshot.k_i);
 986         root_snapshot.k.p.offset = U32_MAX;
 987         root_snapshot.v.flags   = 0;
 988         root_snapshot.v.parent  = 0;
 989         root_snapshot.v.subvol  = BCACHEFS_ROOT_SUBVOL;
 990         root_snapshot.v.pad     = 0;
 991         SET_BCH_SNAPSHOT_SUBVOL(&root_snapshot.v, true);
 992
 993         ret = bch2_btree_insert(c, BTREE_ID_snapshots,
 994                                 &root_snapshot.k_i,
 995                                 NULL, NULL, 0);
 996         if (ret)
 997                 return ret;
 998
 999
1000         bkey_subvolume_init(&root_volume.k_i);
1001         root_volume.k.p.offset = BCACHEFS_ROOT_SUBVOL;
1002         root_volume.v.flags     = 0;
1003         root_volume.v.snapshot  = cpu_to_le32(U32_MAX);
1004         root_volume.v.inode     = cpu_to_le64(BCACHEFS_ROOT_INO);
1005
1006         ret = bch2_btree_insert(c, BTREE_ID_subvolumes,
1007                                 &root_volume.k_i,
1008                                 NULL, NULL, 0);
1009         if (ret)
1010                 return ret;
1011
1012         return 0;
1013 }
1014
1015 static int bch2_fs_upgrade_for_subvolumes(struct btree_trans *trans)
1016 {
1017         struct bch_fs *c = trans->c;
1018         struct btree_iter iter;
1019         struct bkey_s_c k;
1020         struct bch_inode_unpacked inode;
1021         int ret;
1022
1023         bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
1024                              SPOS(0, BCACHEFS_ROOT_INO, U32_MAX), 0);
1025         k = bch2_btree_iter_peek_slot(&iter);
1026         ret = bkey_err(k);
1027         if (ret)
1028                 goto err;
1029
1030         if (!bkey_is_inode(k.k)) {
1031                 bch_err(c, "root inode not found");
1032                 ret = -ENOENT;
1033                 goto err;
1034         }
1035
1036         ret = bch2_inode_unpack(k, &inode);
1037         BUG_ON(ret);
1038
1039         inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
1040
1041         ret = bch2_inode_write(trans, &iter, &inode);
1042 err:
1043         bch2_trans_iter_exit(trans, &iter);
1044         return ret;
1045 }
1046
1047 int bch2_fs_recovery(struct bch_fs *c)
1048 {
1049         const char *err = "cannot allocate memory";
1050         struct bch_sb_field_clean *clean = NULL;
1051         struct jset *last_journal_entry = NULL;
1052         u64 blacklist_seq, journal_seq;
1053         bool write_sb = false;
1054         int ret = 0;
1055
1056         if (c->sb.clean)
1057                 clean = read_superblock_clean(c);
1058         ret = PTR_ERR_OR_ZERO(clean);
1059         if (ret)
1060                 goto err;
1061
1062         if (c->sb.clean)
1063                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
1064                          le64_to_cpu(clean->journal_seq));
1065         else
1066                 bch_info(c, "recovering from unclean shutdown");
1067
1068         if (!(c->sb.features & (1ULL << BCH_FEATURE_new_extent_overwrite))) {
1069                 bch_err(c, "feature new_extent_overwrite not set, filesystem no longer supported");
1070                 ret = -EINVAL;
1071                 goto err;
1072         }
1073
1074         if (!c->sb.clean &&
1075             !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
1076                 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
1077                 ret = -EINVAL;
1078                 goto err;
1079         }
1080
1081         if (!(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done))) {
1082                 bch_err(c, "filesystem may have incompatible bkey formats; run fsck from the compat branch to fix");
1083                 ret = -EINVAL;
1084                 goto err;
1085         }
1086
1087         if (!(c->sb.features & (1ULL << BCH_FEATURE_alloc_v2))) {
1088                 bch_info(c, "alloc_v2 feature bit not set, fsck required");
1089                 c->opts.fsck = true;
1090                 c->opts.fix_errors = FSCK_OPT_YES;
1091         }
1092
1093         if (!c->replicas.entries ||
1094             c->opts.rebuild_replicas) {
1095                 bch_info(c, "building replicas info");
1096                 set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
1097         }
1098
1099         if (!c->opts.nochanges) {
1100                 if (c->sb.version < bcachefs_metadata_version_inode_backpointers) {
1101                         bch_info(c, "version prior to inode backpointers, upgrade and fsck required");
1102                         c->opts.version_upgrade = true;
1103                         c->opts.fsck            = true;
1104                         c->opts.fix_errors      = FSCK_OPT_YES;
1105                 } else if (c->sb.version < bcachefs_metadata_version_subvol_dirent) {
1106                         bch_info(c, "filesystem version is prior to subvol_dirent - upgrading");
1107                         c->opts.version_upgrade = true;
1108                         c->opts.fsck            = true;
1109                 } else if (c->sb.version < bcachefs_metadata_version_inode_v2) {
1110                         bch_info(c, "filesystem version is prior to inode_v2 - upgrading");
1111                         c->opts.version_upgrade = true;
1112                 }
1113         }
1114
1115         ret = bch2_blacklist_table_initialize(c);
1116         if (ret) {
1117                 bch_err(c, "error initializing blacklist table");
1118                 goto err;
1119         }
1120
1121         if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) {
1122                 struct journal_replay *i;
1123
1124                 ret = bch2_journal_read(c, &c->journal_entries,
1125                                         &blacklist_seq, &journal_seq);
1126                 if (ret)
1127                         goto err;
1128
1129                 list_for_each_entry_reverse(i, &c->journal_entries, list)
1130                         if (!i->ignore) {
1131                                 last_journal_entry = &i->j;
1132                                 break;
1133                         }
1134
1135                 if (mustfix_fsck_err_on(c->sb.clean &&
1136                                         last_journal_entry &&
1137                                         !journal_entry_empty(last_journal_entry), c,
1138                                 "filesystem marked clean but journal not empty")) {
1139                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1140                         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1141                         c->sb.clean = false;
1142                 }
1143
1144                 if (!last_journal_entry) {
1145                         fsck_err_on(!c->sb.clean, c, "no journal entries found");
1146                         goto use_clean;
1147                 }
1148
1149                 c->journal_keys = journal_keys_sort(&c->journal_entries);
1150                 if (!c->journal_keys.d) {
1151                         ret = -ENOMEM;
1152                         goto err;
1153                 }
1154
1155                 if (c->sb.clean && last_journal_entry) {
1156                         ret = verify_superblock_clean(c, &clean,
1157                                                       last_journal_entry);
1158                         if (ret)
1159                                 goto err;
1160                 }
1161         } else {
1162 use_clean:
1163                 if (!clean) {
1164                         bch_err(c, "no superblock clean section found");
1165                         ret = BCH_FSCK_REPAIR_IMPOSSIBLE;
1166                         goto err;
1167
1168                 }
1169                 blacklist_seq = journal_seq = le64_to_cpu(clean->journal_seq) + 1;
1170         }
1171
1172         if (c->opts.reconstruct_alloc) {
1173                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1174                 drop_alloc_keys(&c->journal_keys);
1175         }
1176
1177         zero_out_btree_mem_ptr(&c->journal_keys);
1178
1179         ret = journal_replay_early(c, clean, &c->journal_entries);
1180         if (ret)
1181                 goto err;
1182
1183         if (blacklist_seq != journal_seq) {
1184                 ret = bch2_journal_seq_blacklist_add(c,
1185                                         blacklist_seq, journal_seq);
1186                 if (ret) {
1187                         bch_err(c, "error creating new journal seq blacklist entry");
1188                         goto err;
1189                 }
1190         }
1191
1192         ret = bch2_fs_journal_start(&c->journal, journal_seq,
1193                                     &c->journal_entries);
1194         if (ret)
1195                 goto err;
1196
1197         ret = read_btree_roots(c);
1198         if (ret)
1199                 goto err;
1200
1201         bch_verbose(c, "starting alloc read");
1202         err = "error reading allocation information";
1203         ret = bch2_alloc_read(c);
1204         if (ret)
1205                 goto err;
1206         bch_verbose(c, "alloc read done");
1207
1208         bch_verbose(c, "starting stripes_read");
1209         err = "error reading stripes";
1210         ret = bch2_stripes_read(c);
1211         if (ret)
1212                 goto err;
1213         bch_verbose(c, "stripes_read done");
1214
1215         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
1216
1217         if (c->opts.fsck ||
1218             !(c->sb.compat & (1ULL << BCH_COMPAT_alloc_info)) ||
1219             !(c->sb.compat & (1ULL << BCH_COMPAT_alloc_metadata)) ||
1220             test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags)) {
1221                 bool metadata_only = c->opts.norecovery;
1222
1223                 bch_info(c, "starting mark and sweep");
1224                 err = "error in mark and sweep";
1225                 ret = bch2_gc(c, true, metadata_only);
1226                 if (ret)
1227                         goto err;
1228                 bch_verbose(c, "mark and sweep done");
1229         }
1230
1231         bch2_stripes_heap_start(c);
1232
1233         clear_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
1234         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1235
1236         /*
1237          * Skip past versions that might have possibly been used (as nonces),
1238          * but hadn't had their pointers written:
1239          */
1240         if (c->sb.encryption_type && !c->sb.clean)
1241                 atomic64_add(1 << 16, &c->key_version);
1242
1243         if (c->opts.norecovery)
1244                 goto out;
1245
1246         bch_verbose(c, "starting journal replay");
1247         err = "journal replay failed";
1248         ret = bch2_journal_replay(c);
1249         if (ret)
1250                 goto err;
1251         bch_verbose(c, "journal replay done");
1252
1253         if (test_bit(BCH_FS_NEED_ALLOC_WRITE, &c->flags) &&
1254             !c->opts.nochanges) {
1255                 /*
1256                  * note that even when filesystem was clean there might be work
1257                  * to do here, if we ran gc (because of fsck) which recalculated
1258                  * oldest_gen:
1259                  */
1260                 bch_verbose(c, "writing allocation info");
1261                 err = "error writing out alloc info";
1262                 ret = bch2_alloc_write_all(c, BTREE_INSERT_LAZY_RW);
1263                 if (ret) {
1264                         bch_err(c, "error writing alloc info");
1265                         goto err;
1266                 }
1267                 bch_verbose(c, "alloc write done");
1268         }
1269
1270         if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1271                 bch2_fs_lazy_rw(c);
1272
1273                 err = "error creating root snapshot node";
1274                 ret = bch2_fs_initialize_subvolumes(c);
1275                 if (ret)
1276                         goto err;
1277         }
1278
1279         bch_verbose(c, "reading snapshots table");
1280         err = "error reading snapshots table";
1281         ret = bch2_fs_snapshots_start(c);
1282         if (ret)
1283                 goto err;
1284         bch_verbose(c, "reading snapshots done");
1285
1286         if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1287                 /* set bi_subvol on root inode */
1288                 err = "error upgrade root inode for subvolumes";
1289                 ret = bch2_trans_do(c, NULL, NULL, BTREE_INSERT_LAZY_RW,
1290                                     bch2_fs_upgrade_for_subvolumes(&trans));
1291                 if (ret)
1292                         goto err;
1293         }
1294
1295         if (c->opts.fsck) {
1296                 bch_info(c, "starting fsck");
1297                 err = "error in fsck";
1298                 ret = bch2_fsck_full(c);
1299                 if (ret)
1300                         goto err;
1301                 bch_verbose(c, "fsck done");
1302         } else if (!c->sb.clean) {
1303                 bch_verbose(c, "checking for deleted inodes");
1304                 err = "error in recovery";
1305                 ret = bch2_fsck_walk_inodes_only(c);
1306                 if (ret)
1307                         goto err;
1308                 bch_verbose(c, "check inodes done");
1309         }
1310
1311         if (enabled_qtypes(c)) {
1312                 bch_verbose(c, "reading quotas");
1313                 ret = bch2_fs_quota_read(c);
1314                 if (ret)
1315                         goto err;
1316                 bch_verbose(c, "quotas done");
1317         }
1318
1319         mutex_lock(&c->sb_lock);
1320         /*
1321          * With journal replay done, we can clear the journal seq blacklist
1322          * table:
1323          */
1324         BUG_ON(!test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags));
1325         if (le16_to_cpu(c->sb.version_min) >= bcachefs_metadata_version_btree_ptr_sectors_written)
1326                 bch2_sb_resize_journal_seq_blacklist(&c->disk_sb, 0);
1327
1328         if (c->opts.version_upgrade) {
1329                 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1330                 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1331                 write_sb = true;
1332         }
1333
1334         if (!test_bit(BCH_FS_ERROR, &c->flags)) {
1335                 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
1336                 write_sb = true;
1337         }
1338
1339         if (c->opts.fsck &&
1340             !test_bit(BCH_FS_ERROR, &c->flags) &&
1341             !test_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags)) {
1342                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
1343                 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 0);
1344                 write_sb = true;
1345         }
1346
1347         if (write_sb)
1348                 bch2_write_super(c);
1349         mutex_unlock(&c->sb_lock);
1350
1351         if (!(c->sb.compat & (1ULL << BCH_COMPAT_extents_above_btree_updates_done)) ||
1352             !(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done)) ||
1353             le16_to_cpu(c->sb.version_min) < bcachefs_metadata_version_btree_ptr_sectors_written) {
1354                 struct bch_move_stats stats;
1355
1356                 bch_move_stats_init(&stats, "recovery");
1357
1358                 bch_info(c, "scanning for old btree nodes");
1359                 ret = bch2_fs_read_write(c);
1360                 if (ret)
1361                         goto err;
1362
1363                 ret = bch2_scan_old_btree_nodes(c, &stats);
1364                 if (ret)
1365                         goto err;
1366                 bch_info(c, "scanning for old btree nodes done");
1367         }
1368
1369         ret = 0;
1370 out:
1371         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1372         bch2_flush_fsck_errs(c);
1373
1374         if (!c->opts.keep_journal) {
1375                 bch2_journal_keys_free(&c->journal_keys);
1376                 bch2_journal_entries_free(&c->journal_entries);
1377         }
1378         kfree(clean);
1379         if (ret)
1380                 bch_err(c, "Error in recovery: %s (%i)", err, ret);
1381         else
1382                 bch_verbose(c, "ret %i", ret);
1383         return ret;
1384 err:
1385 fsck_err:
1386         bch2_fs_emergency_read_only(c);
1387         goto out;
1388 }
1389
1390 int bch2_fs_initialize(struct bch_fs *c)
1391 {
1392         struct bch_inode_unpacked root_inode, lostfound_inode;
1393         struct bkey_inode_buf packed_inode;
1394         struct qstr lostfound = QSTR("lost+found");
1395         const char *err = "cannot allocate memory";
1396         struct bch_dev *ca;
1397         LIST_HEAD(journal);
1398         unsigned i;
1399         int ret;
1400
1401         bch_notice(c, "initializing new filesystem");
1402
1403         mutex_lock(&c->sb_lock);
1404         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
1405         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
1406
1407         if (c->opts.version_upgrade) {
1408                 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1409                 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1410                 bch2_write_super(c);
1411         }
1412         mutex_unlock(&c->sb_lock);
1413
1414         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
1415         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1416
1417         for (i = 0; i < BTREE_ID_NR; i++)
1418                 bch2_btree_root_alloc(c, i);
1419
1420         set_bit(BCH_FS_ALLOC_REPLAY_DONE, &c->flags);
1421         set_bit(BCH_FS_BTREE_INTERIOR_REPLAY_DONE, &c->flags);
1422         set_bit(JOURNAL_RECLAIM_STARTED, &c->journal.flags);
1423
1424         err = "unable to allocate journal buckets";
1425         for_each_online_member(ca, c, i) {
1426                 ret = bch2_dev_journal_alloc(ca);
1427                 if (ret) {
1428                         percpu_ref_put(&ca->io_ref);
1429                         goto err;
1430                 }
1431         }
1432
1433         /*
1434          * journal_res_get() will crash if called before this has
1435          * set up the journal.pin FIFO and journal.cur pointer:
1436          */
1437         bch2_fs_journal_start(&c->journal, 1, &journal);
1438         bch2_journal_set_replay_done(&c->journal);
1439
1440         err = "error going read-write";
1441         ret = bch2_fs_read_write_early(c);
1442         if (ret)
1443                 goto err;
1444
1445         /*
1446          * Write out the superblock and journal buckets, now that we can do
1447          * btree updates
1448          */
1449         err = "error marking superblock and journal";
1450         for_each_member_device(ca, c, i) {
1451                 ret = bch2_trans_mark_dev_sb(c, ca);
1452                 if (ret) {
1453                         percpu_ref_put(&ca->ref);
1454                         goto err;
1455                 }
1456
1457                 ca->new_fs_bucket_idx = 0;
1458         }
1459
1460         err = "error creating root snapshot node";
1461         ret = bch2_fs_initialize_subvolumes(c);
1462         if (ret)
1463                 goto err;
1464
1465         bch_verbose(c, "reading snapshots table");
1466         err = "error reading snapshots table";
1467         ret = bch2_fs_snapshots_start(c);
1468         if (ret)
1469                 goto err;
1470         bch_verbose(c, "reading snapshots done");
1471
1472         bch2_inode_init(c, &root_inode, 0, 0,
1473                         S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);
1474         root_inode.bi_inum      = BCACHEFS_ROOT_INO;
1475         root_inode.bi_subvol    = BCACHEFS_ROOT_SUBVOL;
1476         bch2_inode_pack(c, &packed_inode, &root_inode);
1477         packed_inode.inode.k.p.snapshot = U32_MAX;
1478
1479         err = "error creating root directory";
1480         ret = bch2_btree_insert(c, BTREE_ID_inodes,
1481                                 &packed_inode.inode.k_i,
1482                                 NULL, NULL, 0);
1483         if (ret)
1484                 goto err;
1485
1486         bch2_inode_init_early(c, &lostfound_inode);
1487
1488         err = "error creating lost+found";
1489         ret = bch2_trans_do(c, NULL, NULL, 0,
1490                 bch2_create_trans(&trans,
1491                                   BCACHEFS_ROOT_SUBVOL_INUM,
1492                                   &root_inode, &lostfound_inode,
1493                                   &lostfound,
1494                                   0, 0, S_IFDIR|0700, 0,
1495                                   NULL, NULL, (subvol_inum) { 0 }, 0));
1496         if (ret) {
1497                 bch_err(c, "error creating lost+found");
1498                 goto err;
1499         }
1500
1501         if (enabled_qtypes(c)) {
1502                 ret = bch2_fs_quota_read(c);
1503                 if (ret)
1504                         goto err;
1505         }
1506
1507         err = "error writing first journal entry";
1508         ret = bch2_journal_flush(&c->journal);
1509         if (ret)
1510                 goto err;
1511
1512         mutex_lock(&c->sb_lock);
1513         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1514         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1515
1516         bch2_write_super(c);
1517         mutex_unlock(&c->sb_lock);
1518
1519         return 0;
1520 err:
1521         pr_err("Error initializing new filesystem: %s (%i)", err, ret);
1522         return ret;
1523 }